I’m trying to get it to work on an Azure VM, but it’s not recognising the GPU
The nVidia GPU extension is installed.
nvidia-smi seems to indicate that the GPU is properly connected or recognised by the VM.
It’s the Standard NC40ads H100 v5 (40 vcpus, 320 GiB memory) deployment.
So an H100 machine, with an AMD CPU with 96 Cores and 320GB RAM running Windows 10.
Any ideas of what I can try to get it working?
Thanks!
I think you need to run TVAI via command line instead of the User interface.
Next Question is, as far as i know there are no Windows drivers for the big Nvidia GPUs line the H100, but try the command line.
To use TVAI via the userinterface the GPU does need to support DirectX.
Thanks, I wasn’t aware of the TVAI CLI.
Before I provisioned the nvidia GPU extension through the Azure interface as per screenshot 2 in the first post, I did try downloading it manually and it seems there is a Windows driver
I did try the PCIe version though, not the NVL version one though, so not sure if that makes a difference.
I’m very new to GPU CLI jobs etc. but will have a read of the docs linked and give it a go.
Thanks for the lead TPX!
Doesn’t seem like the Starlight mini model is listed in the models with CLI list 
The Azure NC40ads H100 v5 VM uses an NVIDIA H100 data-center card, and NVIDIA only ships a compute-mode (TCC) driver for it—there is no WDDM graphics driver. Because of that, Windows treats the H100 purely as a headless compute device rather than a display adapter, so any software that checks for a DirectX/OpenGL-capable GPU reports “no compatible GPU found.” CUDA, PyTorch, or other compute workloads run fine (ideally under Linux or WSL 2), but desktop graphics, video-upsampling tools, or game engines will not. If you need a GPU that Windows apps can see, you must choose a VM with a workstation-class card such as an A10, L40, A40, or T4 instead.
I see. How unfortunate.
Does that mean Starlight would never be an option with H100 (or B200?) even via the CLI?
What would be the recommended best possible cards from Azure then from what you listed?
No, on an H100 (or upcoming B200) the normal Topaz Video AI build can’t see the card, even from the CLI, because Windows never exposes a WDDM adapter. Starlight will only run there if you license Topaz’s enterprise Docker image under Linux. For an Azure VM that “just works” with the desktop license, pick something with a workstation-class WDDM driver: the NVads A10 v5 is the fastest widely available choice, while the NCas T4 v3 works if you need to save cost.
Awesome, thank you mate. I’ll give that a go 
officially Starlight mini is not CLI ready yet, inofficially I’m sure it works, but something is missing with my test, maybe path to the model? CLI & Starlight mini
%ProgramData%\Topaz Labs LLC\Topaz Video AI\models\runner.exe sample_source.mp4 sample_out.mp4 --start-frame-idx 0 --end-frame-idx 64 --ffmpeg-encoding "-c:v h264_nvenc -profile:v high -pix_fmt yuv420p -g 15 -preset p7 -tune hq -rc constqp -qp 18 -rc-lookahead 10 -spatial_aq 1 -aq-strength 15 -b:v 0 -movflags frag_keyframe+empty_moov+delay_moov+use_metadata_tags+write_colr -bf 0" --ffmpeg-preproc-filters scale=w=640:h=480,setsar=1 --upscale-factor 3 --**max-gpu-mem 16**